高温物体红外双目测量及精度研究-research on infrared binocular measurement and accuracy of high temperature objects.docxVIP

the eight neighborhood pixels correlation, redesign the hidden layer considering the nonlinear correlation factors. The experimental results show that infrared image inhomogeneous de gree have decreased to 4.986% with the improved BP neural network correction method, which can obtain better corrected results than other algorithms. Through analyzing the characteristics of infrared image, this paper discusses the platform histogram infrared reinforcement method, high-pass filtering infrared image enhancement method and weighted average infrared image enhancement method combining the infrared image quality evaluation inde x with doing some experiments.We have designed the infrared visual ball targets especially according to the infrared imaging mecha nism combining with the features of infrared image and put forward the new infrared binocular measurement system calibrating scheme through high precision electronic three-dimensional translational platform. Through exploring the comparison between the thermal image calibrating methods, the calibration of the measurement system is completed using radial consistent constraint criterion combining with the LM parameter optimization algorithm.Through studying the existing image edge extraction algorithm, combining with the advantages of neural network and fuzzy reasoning method towards the infrared image characteristics, the left and right correlation gradient is introd uced and infrared image edge extraction method based on eight pixel point fuzzy neural network is proposed. This method can make the edge single-pixel and improve the immun ity ability. Infrared binocular vision feature matching is analyzed in this pape r, and the size measurement is completed with Hough transformation using SURF operator feature matching. The experimental results show that the system size measurement relative error is about 1%, and the system has good stability.Keywords: High-temperature objects；Infrared image；Binocular calibration；Edge detec